Marvell Unveils 2nm Optical Interconnect Tech for AI Data Centers

Marvell pushes optical connectivity to 2nm for AI infrastructure

Marvell Technology has announced a series of industry-first optical interconnect demonstrations built on 2-nanometer process technology, targeting the escalating bandwidth and power demands of AI data centers. The semiconductor company will showcase the technologies at the European Conference on Optical Communication (ECOC) in Málaga, Spain, from September 20–24, 2026, according to details first reported by Marvell.

The demonstrations include three firsts at the 2nm node: a 400G-per-lane optical PAM4 technology, an 800G ZR/ZR+ pluggable with integrated media access control security (MACsec), and 1.6T ZR and O-band coherent-lite technologies. Marvell will also demonstrate its 102.4 Tbps co-packaged optics (CPO) platform, which integrates optics directly with AI accelerators and switch silicon using 200G-per-lane silicon photonics.

Why it matters

As AI data center architectures scale from 1.6 terabits to 3.2 terabits and beyond, operators face a critical challenge: delivering higher bandwidth while reducing power consumption per bit. Optical interconnects at advanced process nodes like 2nm directly address this bottleneck, enabling faster communication between processors, accelerators, and memory while improving energy efficiency. The integration of security features like MACsec at the physical layer also addresses growing concerns about data protection in hyperscale deployments.

Building on multi-generation optical leadership

The 2nm demonstrations extend Marvell's track record of process node leadership in optical connectivity. The company previously introduced the first 5nm 200G-per-lane 1.6T Nova DSPs in 2023, followed by the 3nm 1.6T Ara platform in 2024 and 2nm 800G Libra DSPs earlier this year.

The new 2nm 400G-per-lane optical PAM4 technology is designed to enable the transition to 3.2T data center connectivity, delivering low-power, ultra-high-speed optical connections for AI infrastructure scaling. The 800G ZR/ZR+ OSFP pluggable, powered by the Marvell Libra 2nm coherent DSP, adds MACsec integration to expand the company's coherent DSP and COLORZ pluggable portfolios for secure optical transmission in hyperscale AI and cloud networks.

The 1.6T ZR and 1.6T coherent-lite demonstrations utilize next-generation Marvell DSPs optimized for high-bandwidth connectivity within and between data center facilities, paving the way for 3.2T connectivity.

Co-packaged optics for AI fabric scaling

Marvell's CPO platform addresses a fundamental limitation as per-lane speeds increase and multi-rack AI pods expand beyond copper's reach. By integrating optics directly with AI accelerators and switch silicon, the platform delivers the bandwidth density and power efficiency required to scale AI fabrics. The 102.4 Tbps demonstration with 200G-per-lane silicon photonics represents a significant step toward addressing the interconnect challenges of next-generation AI clusters.

"AI data center operators need optical interconnect technologies that deliver higher bandwidth, lower power and greater security across every part of the network," said Xi Wang, senior vice president and general manager of Marvell's Connectivity Business Unit.

At ECOC 2026, Marvell will present 38 advanced technology demonstrations spanning its AI infrastructure portfolio, including 224G SerDes, 1.6T optics, coherent interconnects, and optical memory fabrics. The company will also participate in presentations and panels on optical connectivity for next-generation AI data center infrastructure.

Marvell first announced these demonstrations in a press release distributed through Business Wire.